Concepts

Audit & provenance

Every quote is re-derivable. Audit blob, provenance chain, conformal intervals, immutability.

ARCNM quotes are defensible. Every cost field traces back to:

Which input the feature came from (3D geometry, drawing, or both).
Which process the time estimate was computed for (milling, turning, sheet-metal forming, etc.).
Which source supplied the rate (manual override, parent environment, environment calibration, platform default).
Which calibration run produced the active coefficients, on what evidence, at what timestamp.

When a procurement team asks "how did this part get to €12.84?", the audit blob answers in 30 seconds, not a week.

The audit blob

GET /api/v1/parts/calculations/{id} returns the result + a deep analytics object. The audit blob lives at:

analytics.cost_pipeline

with one section per stage that ran (extraction, drawing, fusion, planning, physics, cost_decomposition).

The analytics blob is free-form and its internal field names are illustrative — treat it as informational and don't hardcode the exact shape. The stable top-level fields (unit_cost, total_cost, setup_cost, currency, status, timestamps) are documented in the Calculations reference.

Top-level structure

{
  "id": "9d3f…",
  "status": "succeeded",
  "unit_cost": 12.84,
  "currency": "EUR",
  "analytics": {
    "cost_pipeline": {
      "engine_version":  "2026-05-21",
      "schema_version":  "v2",
      "wallclock_s":     32.4,
      "stages": {
        "extraction":         { /* … */ },
        "drawing":            { /* … */ },
        "fusion":             { /* … */ },
        "planning":           { /* … */ },
        "physics":            { /* … */ },
        "cost_decomposition": { /* … */ }
      },
      "provenance": { /* per-parameter resolved-source map */ },
      "intervals":  { /* conformal intervals at each stage */ }
    }
  }
}

Per-stage extracts

Extraction (`stages.extraction`)

{
  "envelope_mm":  { /* … */ },
  "features":     [ /* … */ ],
  "dfm_issues":   [ /* … */ ],
  "material_classification": { "iso_group": "P" },
  "wallclock_s":  4.1
}

Drawing (`stages.drawing`)

Carries the typed DrawingExtraction + per-field confidence + validation report.

{
  "extraction_tier":          "arcnm",
  "extraction_tier_display":  "ARCNM",
  "wallclock_s":              21.0,
  "drawing_extraction":       { /* typed DrawingExtraction */ },
  "validation_report": [
    {
      "rule_id":          "R-ISO286-001",
      "severity":         "confidence_adjust",
      "field_path":       "pmi_supplements[3]",
      "message":          "Tolerance pair (+25.0/-0.0 µm) disagrees with ISO 286 lookup (+21/0 µm) for j6.",
      "value":            [25.0, 0.0],
      "expected":         [21, 0],
      "clause":           "ISO 286-1 §A",
      "confidence_factor": 0.5
    }
  ],
  "self_correction": [ /* per-field retry log */ ]
}

Validation rule_ids are stable: R-ISO286-001/002, R-ISO2768-001/002, R-Y14_5-001..010, R-TB-001/002/003, R-CHM-001, R-RAD-001, R-Ra-001/002/003, R-THD-001/002, R-MAT-001. Switch on them in your downstream pipelines.

Physics (`stages.physics`)

For each operation, the cost inputs the engine consumed (each resolved through the environment cascade) and the times produced:

{
  "operations": [
    {
      "operation_id":  "op_a8…",
      "process":       "drilling",
      "feature_id":    "feat_92…",
      "inputs": {
        "tool_diameter_mm":          8.0,
        "feed_rate_mm_per_rev":      { "value": 0.18, "source": "platform_default" },
        "cutting_speed_m_per_min":   { "value": 35.0, "source": "environment_override" }
      },
      "outputs": {
        "setup_time_s":   42.0,
        "cycle_time_s":   18.5,
        "tool_life_units": 850
      }
    }
  ]
}

Cost decomposition (`stages.cost_decomposition`)

{
  "decomposition": { /* per-line-item costs — see Cost concepts */ },
  "lot_size_curve": { /* points across batch sizes */ }
}

The conservative tenant-specific adjustment that calibration applies is recorded here too, with its supporting evidence, so every quote stays auditable.

Provenance map

analytics.cost_pipeline.provenance is a flat map from parameter path to the source that resolved it:

{
  "milling.feed_rate_mm_per_rev":      "environment_calibration",
  "milling.cutting_speed_m_per_min":   "environment_override",
  "labour_rate_eur_per_hour":          "platform_default",
  "machine_rate_eur_per_hour.haas_vf2":"environment_override",
  "material.steel.1.0577.eur_per_kg":  "environment_catalogue"
}

Run jq on it to assert at deploy time:

arc calc get $CALC | jq '.analytics.cost_pipeline.provenance | with_entries(select(.value == "platform_default"))'
# Empty list = every parameter is calibrated, no defaults leaking in.

Immutability

Every successful calculation row is append-only:

The audit blob is stored as-is in versioned, lifecycle-managed object storage.
The row's analytics field is read-only after the worker writes it.
Re-running the same part_revision_id + costing_environment_id + lot_size + material is not idempotent — it creates a new Calculation row with a new id, so calibration changes are visible.
The previous row stays accessible by id forever.

For "the price as it was on 2026-05-28" queries, persist the calculation_id in your ERP and re-fetch from us — we keep them.

Conformal intervals

Three levels of interval:

Level	Field	Width depends on
Per-parameter	`inputs.<key>.interval`	Calibration sample count + spread
Per-operation	`outputs.<key>.interval`	Propagated from the input intervals it depends on
Quote-level	`unit_cost_interval_eur`	Combined platform + environment evidence (conservative)

Where platform and environment evidence both apply, ARCNM reports a conservative interval rather than the tightest one it could — by design (see Calibration → Conformal intervals).

Compliance & retention

HGB §257 / GoBD: 10-year retention on audit blobs for German customers.
GDPR: no personal data in the audit blob by construction; only CAD-derived geometry and pricing math.
SOC 2: audit entries are part of the SOC 2 control set. Read them via the audit API with the audit:read scope.

Exporting to your data warehouse

The audit API returns audit rows with cursor pagination — page through them and drop into Snowflake / BigQuery / Postgres for cost analytics. Requires the audit:read scope; narrow the stream with the available action and resource-type filters.